A new demo from DS Stream shows how the Product Intelligence engine in our pricing optimization tool helps retailers price products that have no sales history yet. It runs under six minutes and is meant for pricing and category teams at multi-category retailers who bring in hundreds or thousands of new SKUs every season.
Watch on YouTube: Price New Products Smarter with AI Product Intelligence
The cold start problem with new products
A new SKU arrives without any historical sales or inventory data, so there is nothing to base its price on. The demo uses a running shoe retailer launching a new model with Gore-Tex as its example. Without past data, two things tend to happen. First, the legacy system files the new model under one of the old categories and applies an arbitrary target margin that may not fit this type of product at all. Second, cannibalization goes unnoticed. A discount on regular running shoes can pull sales away from premium models that customers see as identical substitutes, and the team has no way of seeing it.
Product Intelligence starts from the product itself. Instead of relying on the ERP taxonomy, it processes the catalog, reads each product description and catalog photo, and looks for similarities between new items and products that already have a sales record.
What the demo shows
- Prices based on sister models. The engine maps the new shoe's visual features and attributes to the historical demand of its sister models. In the demo, pricing the shoe at the profit-optimal point on day one gives a 4.1% higher launch margin compared with its original price.
- A closest substitutes view. The engine finds that the new trail shoe has a 93% visual and technical similarity match with two existing high-margin styles. With that link visible, the team can stop a planned discount that would have cannibalized its newly launched flagship line.
- Product families across the catalog. In the demo, about 50K SKUs are grouped into more than 1K detected product families, with high confidence scores. The features of each SKU come from its description and from image detection.
- Search by technical specs. Category managers can filter the catalog by technical features. Gore-Tex weather styles form one family, for example, while stack cushioning shows up in two.
Traditional retail databases work with SKU numbers and basic categories. The engine works with what the customer notices: visual design, technical fabric specs, and direct substitutes. For a launch, the team learns which existing SKUs it can borrow proven elasticity curves and pricing from, and where a discount could eat into margin elsewhere in the catalog.
Where to go next
This demo picks up where our product segmentation demo left off. That one sorts the existing catalog by how products sell, and this one handles the products that have no selling record yet. The full recording of our Pricing at Scale session is on the DS Stream webinars page. For the method behind both demos, from category rules to SKU-level models, read our guide to retail price optimization.
If you want to see how this would work for your own product launches, see how DS Stream approaches pricing optimization for retailers.
